On January 28, Tencent's Hunyuan team made a significant announcement—the HunyuanImage 3.0 text-to-image model (designated as HunyuanImage 3.0-Instruct) has been made open-source. This advanced model is constructed on a Mixture of Experts architecture boasting 80 billion parameters. It offers robust support for image editing and multi-image fusion functionalities. What sets it apart is its incorporation of a 'think-before-edit' mechanism, which ensures a consistently high quality of output.
Impressively, the model has claimed the seventh position on the LMArena image editing leaderboard, distinguishing itself as the sole open-source model within the top seven echelon. At present, both the model weights and the comprehensive codebase have been made accessible on prominent open-source platforms, including Github and Hugging Face, fostering wider community engagement and innovation.